usb: pipelined register writes for the bring-up; Jaguar3 stage timing - #417

Merged

josephnef merged 29 commits into

OpenIPC:masterfrom

gilankpam:perf/usb-pipelined-init-writes

Sep 17, 2026

Commits on Sep 17, 2026

usb: pipelined register writes for the bring-up; Jaguar3 stage timing

authored and
josephnef
committed
04b216f
usb pipelining: review fixes — shared callback pool, 8822C settle flush, per-transport xfer count

josephnef
and
claude
committed
1627d0f
tests: j3_tx_flood_ab.sh — alternating TX-flood A/B of two trees on one Jaguar3 DUT

josephnef
and
claude
committed
ecce836
usb pipelining: flush on every settle; RF-window check pokes the bits it tests

josephnef
and
claude
committed
47e741c
usb pipelining: failure paths — sync fallback, abandoned-queue short-circuit, one-shot scope

josephnef
and
claude
committed
cc74109
usb pipelining: callback-safe pool accounting; drain before the efuse trigger settle

josephnef
and
claude
committed
796047b
usb pipelining: publish a slot last; drain deadline in elapsed time; harness exit codes

josephnef
and
claude
committed
bb8bbe3
usb pipelining: the destructor waits for a callback still inside a slot; chipstate rejects --no-claim with --init

josephnef
and
claude
committed
04202d7
usb pipelining: a failed completion fails the bring-up; scope the timing claims

josephnef
and
claude
committed
eca3a8a
usb pipelining: reads do not fail the batch; one post-cancel deadline; last-chance reap

josephnef
and
claude
committed
5575081
usb pipelining: the batch verdict survives abandonment; readiness rolls back on a failed close

josephnef
and
claude
committed
fc198f5
usb pipelining: fix the CW-tone path build; flood A/B reports the submitted tally

josephnef
and
claude
committed
e00b65a
tests: j3_tx_flood_ab.sh reports the submitted tally and matches the first-TX stage by key

josephnef
and
claude
committed
f9cf73b
jaguar3: the RF-window write-back covers the whole window; claims carry that scope

josephnef
and
claude
committed
24b195c
jaguar3: stage checkpoints drain the queue first; RF check fails on a failed restore

josephnef
and
claude
committed
6918386
usb pipelining: the destructor keeps a slot seen in flight before waiting out its callback

josephnef
and
claude
committed
f54a196
chipstate: build the raw adapter inside the register-op try block

josephnef
and
claude
committed
3bfa42c
usb pipelining: bulk-IN drains first; the callback's free-list push is its last act

josephnef
and
claude
committed
0a01bd4
usb pipelining: an interrupted event wait is a wake-up; RF check counts its read-back words

josephnef
and
claude
committed
45549a6
usb pipelining: reads fall back synchronously too; retired reads are not write errors

josephnef
and
claude
committed
a632a81
usb: count register transfers where they are issued

josephnef
and
claude
committed
ec81294
tests: j3_rf_window_readback.sh names the probed register in its diagnostics

josephnef
and
claude
committed
f8a6128
usb pipelining: snapshot the completion counter before looking, not after

josephnef
and
claude
committed
77b777c
usb pipelining: the batch opens even when pipelining cannot

josephnef
and
claude
committed
04c0cd2
usb pipelining: a stale completion cannot touch a later batch's verdict

josephnef
and
claude
committed
6275f80
usb pipelining: batches nest; handoff and free decision share the pool mutex; re-init joins coex

josephnef
and
claude
committed
814cb6c
jaguar3: readiness is provisional from the top of InitWrite

josephnef
and
claude
committed
2375a45
jaguar3: attach the transfer counter to the timers on USB only

josephnef
and
claude
committed
dd47acb
jaguar3: refuse InitWrite under a live RX loop; the coex stop flag is atomic

josephnef
and
claude
committed
7ff9657

Navigation Menu

Uh oh!

usb: pipelined register writes for the bring-up; Jaguar3 stage timing - #417

usb: pipelined register writes for the bring-up; Jaguar3 stage timing#417
josephnef merged 29 commits into
OpenIPC:masterfrom
gilankpam:perf/usb-pipelined-init-writes

Commits on Sep 17, 2026

usb: pipelined register writes for the bring-up; Jaguar3 stage timing

usb pipelining: review fixes — shared callback pool, 8822C settle flush, per-transport xfer count

tests: j3_tx_flood_ab.sh — alternating TX-flood A/B of two trees on one Jaguar3 DUT

usb pipelining: flush on every settle; RF-window check pokes the bits it tests

usb pipelining: failure paths — sync fallback, abandoned-queue short-circuit, one-shot scope

usb pipelining: callback-safe pool accounting; drain before the efuse trigger settle

usb pipelining: publish a slot last; drain deadline in elapsed time; harness exit codes

usb pipelining: the destructor waits for a callback still inside a slot; chipstate rejects --no-claim with --init

usb pipelining: a failed completion fails the bring-up; scope the timing claims

usb pipelining: reads do not fail the batch; one post-cancel deadline; last-chance reap

usb pipelining: the batch verdict survives abandonment; readiness rolls back on a failed close

usb pipelining: fix the CW-tone path build; flood A/B reports the submitted tally

tests: j3_tx_flood_ab.sh reports the submitted tally and matches the first-TX stage by key

jaguar3: the RF-window write-back covers the whole window; claims carry that scope

jaguar3: stage checkpoints drain the queue first; RF check fails on a failed restore

usb pipelining: the destructor keeps a slot seen in flight before waiting out its callback

chipstate: build the raw adapter inside the register-op try block

usb pipelining: bulk-IN drains first; the callback's free-list push is its last act

usb pipelining: an interrupted event wait is a wake-up; RF check counts its read-back words

usb pipelining: reads fall back synchronously too; retired reads are not write errors

usb: count register transfers where they are issued

tests: j3_rf_window_readback.sh names the probed register in its diagnostics

usb pipelining: snapshot the completion counter before looking, not after

usb pipelining: the batch opens even when pipelining cannot

usb pipelining: a stale completion cannot touch a later batch's verdict

usb pipelining: batches nest; handoff and free decision share the pool mutex; re-init joins coex

jaguar3: readiness is provisional from the top of InitWrite

jaguar3: attach the transfer counter to the timers on USB only

jaguar3: refuse InitWrite under a live RX loop; the coex stop flag is atomic